Predicting protein–protein interactions in the post synaptic density

YMCNE-02813; No of Pages 12 Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Molecular and Cellular Neuroscience journal homepage: www.elsevier.com/locate/ymcne

Q1 2 3

Predicting protein–protein interactions in the post synaptic density Ossnat Bar-shira ⁎, Gal Chechik

F

1

The Gonda Brain Research Center, Bar-Ilan University, Ramat Gan 52900, Israel

i n f o

a b s t r a c t

Article history: Received 25 January 2012 Revised 9 April 2013 Accepted 19 April 2013 Available online xxxx

R O

a r t i c l e

P

The post synaptic density (PSD) is a specialization of the cytoskeleton at the synaptic junction, composed of hundreds of different proteins. Characterizing the protein components of the PSD and their interactions can help elucidate the mechanism of long-term changes in synaptic plasticity, which underlie learning and memory. Unfortunately, our knowledge of the proteome and interactome of the PSD is still partial and noisy. In this study we describe a computational framework to improve the reconstruction of the PSD network. The approach is based on learning the characteristics of PSD protein interactions from a set of trusted interactions, expanding this set with data collected from large scale repositories, and then predicting novel interaction with proteins that are suspected to reside in the PSD. Using this method we obtained thirty predicted interactions, with more than half of which having supporting evidence in the literature. We discuss in details two of these new interactions, Lrrtm1 with PSD-95 and Src with Capg. The first may take part in a mechanism underlying glutamatergic dysfunction in schizophrenia. The second suggests an alternative mechanism to regulate dendritic spines maturation. © 2013 Published by Elsevier Inc.

D

Keywords: Classification Computational biology Network reconstruction Postsynaptic density Protein–protein interactions

E

5 6 7 8 9 10 12 11 13 14 15 16 17 18 19

O

4

T

36

C

35

Introduction

38 39

Understanding the structure and function of mammalian glutamatergic synapses has been a major focus of molecular neuroscience. Particular attention has been given to the postsynaptic density (PSD), a dense complex of proteins whose function is to detect and respond to neurotransmitter that is released from pre-synaptic terminals. Broadly speaking, one set of these proteins is docked to the cell membrane, forming the “front end” of the post synaptic glutamate signaling cascades. These include AMPA (α-amino-3-hydroxy-5methyl-4-isoxazolepropionic acid) and NMDA (N-methyl-D-aspartic acid) receptors, which convert the chemical signals from presynaptic terminal to electrical signal by allowing an influx of positive ions into the cell (Traynelis et al., 2010). Other proteins serve to constantly modulate these signals through several complex mechanisms. First, the flux of ions allowed into the cell is regulated by tuning the distribution and density of receptors or their subunit composition (synaptic plasticity) (Malenka and Bear, 2004; Matta et al., 2011; Sheng and Jong Kim, 2002). Second, the impact of the ion influx is regulated by tuning the morphology of dendritic spines (structural plasticity) (Bourne and Harris, 2008). Proteins within the PSD play a crucial ole in this regulation. For instance, the kinetics of NMDA receptors can be altered by Src tyrosine kinase (Ali et al., 2001), AMPAR trafficking is regulated in part by Pick1 and Grip1 (Kulangara et al., 2007; Volk

46 47 48 49 50 51 52 53 54 55 56 57 58

Q3 59 60

R

R

44 45

N C O

42 43

U

40 41

E

37

⁎ Corresponding author. Fax: +972 3 535 2185. E-mail address: [email protected] (O. Bar-shira).

20 21 22 23 24 25 26 27 28 29 30 31 32 34 33

et al., 2010), and changes in spine shape and size are mediated by proteins like Cortactin, Actin, Src, Capg, Shank and Homer (Fan et al., 2011; Huang et al., 1997; Sala et al., 2001). These plasticity mechanisms tune the transmission of signals through the synapse using a carefully-orchestrated web of protein interactions. Mapping protein–protein interactions is necessary to gain insight into protein function (Legrain, Wojcik, and Gauthier, 2001), detect molecular pathways (Segal, Wang, and Koller, 2003), or identify potential drug targets (Archakov et al., 2003; Hormozdiari et al., 2010). Charting the protein–protein interactions is particularly important in context of the PSD, where modifications in protein conformation and interactions have been linked to neuropsychiatric and neurodegenerative disorders. Known examples include the association between autism and mutations in PSD proteins such as CNTNs, NRXNs or Shank3 (Bourgeron 2009), and between hypofunction of NMDAR receptors and schizophrenia (Kristiansen et al., 2007; Stephan et al., 2006). A solid knowledgebase of the PSD proteome and interactome has a potential to lead to new targets for treating such disorders. In schizophrenia for example, a first approach to enhance NMDA receptor's activity is to target the NMDAR itself, and multiple compounds were proposed for modulating NMDA receptor activity. Unfortunately, despite continuous efforts, most NMDAR-targeting drugs were found ineffective or induce severe side effects (Kalia et al., 2008). Recovering the pathways that control NMDA receptors can provide a new perspective into this problem. NMDAR is regulated by a collection of PSD kinases, phosphatases, and other molecules through multiple regulatory pathways (Hunt and Castillo, 2012), In this paper, our computationally inferred interactions suggest a pathway by which PSD-95, Lrrtm1 and Neurexin decrease Src induced NMDAR tyrosine phosphorylation. Such

1044-7431/$ – see front matter © 2013 Published by Elsevier Inc. http://dx.doi.org/10.1016/j.mcn.2013.04.004

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 Q4 76 77 78 79 80 81 82 83 84 85 86 87 88 89

134 135

To predict novel interactions between PSD proteins, we follow a three-step procedure, illustrated in Fig. 1A. In the first phase, seed network construction, we construct a PPI network using evidence from high confidence interactions of PSD. In the second phase, network expansion with candidate interactions, we expand the seed network using potential interactions proposed by high throughput experiments. To decide which interactions to add, we grow the network in an iterative process, layer by layer (Fig. 1B). At each iteration, we use the current network to train models of interactions (Fig. 1B(iii)), then rank the candidate interactions (Fig. 1B(iv)) and add the highest confidence interactions to the network (Fig. 1B(v)). These steps are repeated until no more interactions are found. The end result of this phase is an expanded network. In the third phase, network expansion with candidate proteins, we predict de novo interactions between the expanded network and proteins that were experimentally pulled out of the PSD. We apply the same iterative expansion procedure as in the second phase, but here we consider all possible interactions between each candidate protein and the network reconstructed so far. The resulting final network is

115 116 117 118 119 120 121 122 123 124 125 126 127 128 129

136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152

C

113 114

E

111 112

R

109 110

R

107 108

O

105 106

C

103 104

N

101 102

U

99 100

F

Overview of the reconstruction approach

97 98

O

133

96

R O

Results

94 95

P

132

92 93

Fig. 1. Network reconstruction is performed in three iterative steps. (A) Starting with a seed network curated interactions, we consider a set of candidate interactions, which have evidence to connect the proteins in the seed network. We then further consider interactions from candidate proteins that are not known to connect to the seed or expanded network. (B) Expanding a network (by either candidate interactions or proteins) takes place in iterations. (i) The network is initialized to be the seed network. (ii) A set of candidate interactions is proposed. (iii) Classifiers are trained on interactions from the existing network. (iv) The candidate interactions are ranked by the trained classifiers. (v) The most likely interaction are validated and added to the current network.

T

130 131

pathways offer a potential explanation for NMDAR diminished activity, and suggest a set of proteins and their domains of interaction to be examined as possible drug targets. Obtaining the full set of protein interactions in the PSD can also improve our understanding of the interplay between activity (synaptic efficacy) and structure (spine morphology). Dendritic spines change their size and shape in correlation with synaptic activity (Kasai et al., 2003), a transformation that involves changes in actin filaments length and organization (Hotulainen and Hoogenraad, 2010). Actin cytoskeleton regulation was already shown to involve multiple proteins that reside in the PSD, including NMDAR, CaMKII, and GTPases (Rho, Ras), and identifying the full list of protein interactions could reveal additional molecular mechanism connecting synaptic efficacy to structural plasticity. Despite the significant effort to identify the full list of proteins of the PSD and their interactions (Bayés et al., 2010; Collins et al., 2006; Cheng et al., 2006; Fernández et al., 2009; Li et al., 2004; Peng et al., 2004; Yoshimura et al., 2003), the current reconstructions of the PSD networks are likely to be partial and noisy. For instance, a meta analysis of studies that detected PSD proteins shows that only 42% of the proteins were detected in more than one study (Collins et al., 2006). Mapping the protein interactions may be similarly noisy: A survey of small scale PPI studies of the NMDA receptor, found that 41% of the proposed PSD proteins (77 out of 186) had no known interactions with the rest of the network (Pocklington et al., 2006). High-throughput measures of protein–protein interactions (PPI) provide very valuable evidence on protein interactions, but they are also susceptible to under- and over-detection (Qi et al., 2006). This calls for developing methods that can combine evidence from multiple experiments and produce a high confidence reconstruction of the PSD network. Here we describe a computational approach to reconstruct the PSD network based on learning the characteristics of PSD protein interaction, and predicting new interacting pairs. Similar approaches were successfully applied in other PPI networks (Skrabanek et al., 2008). We start with a “seed” network of PSD proteins built from high confidence interactions, and then expand that network repeatedly by adding edges from a list of suspected interactions. Finally, we further expand the network using proteins that are suspected to reside in the PSD, and predict how they interact with the network. The end result of this process is a PSD network with 25% more protein interactions than the initial network.

D

90 91

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

E

2

evaluated using the literature and using additional experimental as- 153 says that were not used during training. 154 Training models of interactions using the high confidence PSD network

155

We start with creating a trusted seed network by collecting PPIs from small scale experiments that report interactions within the mouse PSD (Cho et al., 2007; Dong et al., 1997; Jackson and Nicoll, 2009; Leonard et al., 1998; Leonoudakis et al., 2004; Nishimune et al., 1998; Saglietti et al., 2007; Sato et al., 2008; Schulz et al., 2004; Schwenk et al., 2009; Setou et al., 2002; Silverman et al., 2007; Song et al., 1998a,b; Stegmüller et al., 2003; Terashima et al., 2004; Torres et al., 1998; Torres et al., 2001; Uchino et al., 2006; Von Engelhardt et al., 2010; Wang et al., 2006; Xia et al., 1999), and from a comprehensive study that collected PPI from 190 studies (Pocklington et al., 2006). This set of interactions served as a training set for learning binary classifiers that detect repeating patterns that can be used to predict new interactions. In our supervised learning framework, each protein is represented by a vector of measurements from various sources, which are called ‘features’. We selected features that are expected to be highly correlated in pairs of proteins that interact, and less correlated in noninteracting proteins. Fig. 2 shows examples of three features that follow this pattern for one pair of proteins that are known to interact (Gria2, Gria3, Fig. 2A–C) and one pair of proteins that is believed not to interact (Grb2, Actg1, Fig. 2D–F). Fig. 2A depicts the expression profile of Gria2 and Gria3 across a large compendium of microarray experiments (see Experimental methods), showing that the profiles of the two proteins are highly correlated across the compendium. At the same time, the expression profiles of another pair of genes (Grb2,

156 157

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 Q5 179 180

3

P

R O

O

F

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

181

T

E

D

Fig. 2. Examples of similarity between three features of interacting and non-interacting proteins. (A–C) Features of two proteins that are known to interact: Gria2 and Gria3 which code two subunits of the AMPA receptor. (A) Gene expression profiles are highly correlated across conditions; Pearson correlation ρ > 0.9, P b 10−181.(B) Gene expression across 12 brain structures including the Midbrain, Isocortex (CTX), Striatum (STR), Cerebellum, Thalamus (THA), Olfactory areas (OLF), Cortical subplate, Pallidum, Hippocampal formation, Hypothalamus Pons and Medulla, Pearson correlation ρ > 0.9, P b 10−4. (C) Phylogenetic profiles, Pearson correlation ρ > 0.85, P b 10−28. (D–F) Similar to panels C–E, features of two proteins that do not interact, Actg1 and Grb2. (D) Gene expression profiles, ρ b −0.3, P b 10−15, Pearson correlation. (E) Gene expression across 12 brain regions, Pearson correlation ρ b 0.4, P > 0.1. (F) Phylogenetic profiles, Pearson correlation ρ b −0.6, P b 10−14. Gene expression data was scaled (zero mean, unit variance) for better visualization.

196

Selecting a supervised learner

190 191 192 193

197 Q6 198 199 200 201 202 203 204 205 206 207 208 209 210 211

E

R

R

188 189

N C O

186 187

Various methods have been suggested for predicting protein– protein physical interactions, (Bader et al., 2004; Ben-Hur et al., 2005; Chen and Liu, 2005; Jansen et al., 2003). Here we formulate the task of predicting an interaction as a supervised binary classification problem, and tested two approaches in the family of large margin classifiers. Specifically, we used methods based on support vector machine (Vapnik, 1998), see Experimental methods. The first, “global”, approach was to train a single prediction model for the whole network. In the alternative, “local”, approach, we train a separate model for each protein based on its own interactions with other proteins in the network. Since the local models are trained independently, their scores are then calibrated to a unified scale (see Experimental methods). To evaluate the quality of the two classification approaches, we computed the precision of their predictions on held-out data that was not used during training (see Experimental methods). In particular, given

U

184 185

C

194 195

Actg1) are not positively correlated (Fig. 2D), A similar phenomenon is observed in two other features: gene expression profile across different brain structures (Fig. 2B, E) and phylogenetic profiles (Fig. 2C, F). All three exhibit high correlation between interacting protein, (Fig. 2A–C), and little correlation or sometimes anti-correlation in non-interacting proteins (Fig. 2D-F). This is in agreement with previous findings reporting correlation between transcriptome or co-evaluation profiles of interacting proteins (Ge et al., 2001; Goh et al., 2000). Specifically, we combined information from three datasets: mRNA expression profiles from 471 brain related experiments, sequence signatures (protein domains) from InterPro, and ortholog maps of 99 species (see Experimental methods). We excluded proteins that lacked data from one or more sources, like proteins with genes that were not included on the expression gene-chip. The resulting seed network included 295 interactions between 150 proteins.

182 183

a trained model, we used it to predict interactions between all candidate pairs and ranked the pairs by their interaction score. We then computed the fraction of truly interacting proteins within the top-1,2,3,…,k ranked pairs, known as the precision at top-k. Fig. 3A depicts this precision as a function of k (k = 1,2,…,100) for the two classification approaches. The local SVM approach is consistently more precise for all values of k. This suggests that the characteristics that determine if two proteins interact may change significantly from one protein to another, and that the common approach of learning a single model of interaction for the whole network may not have enough flexibility and modeling power.

212

Evaluating the predictive power of various features

223

We examined three sets of features that were demonstrated to be helpful in predicting PPI interactions: expression, signatures and ortholog information (Bhardwaj et al., 2005; Jansen et al., 2002; Ng et al., 2003). We found that each of the three features was predictive on its own and that combining all three features improved the precision significantly (Fig. 3B). Predictions based solely on expression or orthologs were highly precise at the few top ranked interactions, but worse when more than ten interactions are considered.

224 225

Expanding the network with candidate interactions

232

We expanded the seed network using candidate interactions from IntAct and Mint (Chatr-aryamontri et al., 2007; Hermjakob et al., 2004), two databases that list manually curated interactions. Importantly, we avoided using indirect interactions, and excluded interactions that were originally detected as co-complex n-ary interactions in IntAct. To decide which interactions should be added to the network, we used the calibrated scores of all the local classifiers, and added any interaction whose score exceeded a confidence threshold

233

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

213 214 215 216 217 218 219 220 221 222

226 227 228 229 230 231

234 235 236 237 238 239 240

4

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

Expanding the network with candidate proteins

249 250

We next predicted new interactions to the expanded network bases on a list of 765 proteins from pull down experiments (Bayés et al., 2010; Dosemeci et al., 2007; Fernández et al., 2009) (see Experimental methods). We used the same prediction procedure as above, but this time considering all possible interactions between the 203 proteins in the expanded network and the 765 new candidate proteins (the third layer in Fig. 1A). The resulting network, which we call the final network, is illustrated in Fig. 4B–D. Fig. 4B depicts the full final network, Fig. 4C and D focus on the close neighborhoods of interactions from panel B. In all panels, the newly predicted proteins are shown in red. The top 30 predicted interactions are listed in Table 1, and provided online at http://chechiklab.biu.ac.il/~ossnat/ psd_reconstruction.html. Since no ground truth labels exist for these predictions, we evaluated their accuracy in two ways: by looking at properties of the predicted interactions and by comparing them to evidence from the literature. First, we examined if the predicted interacting proteins tend to be more involved in similar biological functions. Specifically, we used two features that were not used in training: phenotypic profiles (Blake et al., 2009) and additional gene expression assays (Edgar, 2002) (see Experimental methods). Before using these datasets to evaluate our predictions, we tested their usefulness on the seed network and the expanded network. For each of these two datasets, we compared the correlation found in interacting and non-interacting pairs, as determined by the seed and expanded network. Indeed, in both networks, the distribution of phenotypic profile similarity was significantly different between interacting and non-interacting proteins (seed network, P b 10 −18, Wilcoxon, expanded P b 10 −31, Fig. 5A and B). A similar significant difference was observed for gene expression (seed, P b 10 −12, Wilcoxon, expanded, P b 10 −22, Fig. 5D–E). These results show that both phenotypes and expression datasets could indicate interactions between proteins, and we therefore used them to measure the similarity of our predicted interactions. Once again we found a highly significant correlation

262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282

Q2

Validating of the top predicted interactions

286

We focused on the top 30 most likely predictions (Fig. 4B), listed in Table 1, where our estimate of the interaction precision was ~ 0.3 (see Experimental methods). For each prediction we conducted a comprehensive search for relevant papers and homologous interactions. Overall, for 18 of the top 30 predicted interactions we found support in previous literature. Of these 18, 13 interactions had evidence for a direct interaction between homologous proteins or similar molecules (Fig. 4B, circles), and 5 had evidence for indirect interactions involving intermediate proteins (Fig. 4B, triangle). As far as we can tell, the remaining twelve pairs were not previously reported in the literature. Since PPIs are preserved between protein pairs with high sequence (Sharan et al., 2005; Yu et al., 2004) and since synaptic proteins are highly conserved across species (Sakarya et al., 2007), we started by seeking evidence for interactions between homologue proteins in other mammals. Few of the pairs where supported by direct evidence, for example, Dynamin/Dnm1 and protein kinase-C/Prkca (Liu et al., 1994) were shown to interact in rat. Other pairs, such as Grb2 and dynamin-3/dnm3, have less direct evidence. Mammals have three dynamin isoforms with very similar sequences (Van der Bliek, 1999). Grb2 was found to interact with dynamin-1/Dnm1(Grabs, 1997) and dynamin-2/Dnm2 (Papin and Subramaniam, 2004). Interestingly, the interaction between Grb2 and dnm1 is through the PQVPSR motif (Grabs, 1997), this domain appears in 3 out of 4 dynamin-3 isoforms, suggesting that Grb2 and Dnm3 also interact. A third example is the interaction between PSD-95/Dlg4 and LRRTM4. Arbuckle et al. found a weak interaction (mean intensity 2265, ranked 214 out of 290 examined peptides) between one peptide of LRRTM4 (ERSHLVPQTPQKPLI) and the SH3 domain of PSD-95 (Arbuckle et al., 2010). To reinforce this evidence, we tracked the features that we used to predict this interaction. Dlg4 and Lrrtm4 show a significantly high correlation across brain related gene expression experiments (P b 10 −90, ρ > 0.7, Pearson correlation, Fig. 6A) and ortholog maps (P b 10 −11, ρ > 0.6, Pearson correlation, Fig. 6B). What about predicted interactions that have no literature evidence? While some are likely to be completely new, it is also possible that we predict interactions between pairs that are actually connected through a third protein. We found five pairs that our classifiers predicted as

287 288

D

T

C

E

260

Q7 261

R

258 259

R

257

O

255 256

C

253 254

N

251 252

U

245 246

R O

248

243 244

between phenotypic profile similarity and the interaction probabili- 283 ties (P b 10 −135, GSEA; P b 10 −100, Spearman, Fig. 5C), and for gene 284 expression (P b 10 −154, GSEA; P b 10 −125, Spearman, Fig. 5F). 285

P

247

(see Experimental methods). We repeated this expansion process with the newly expanded network until no new interactions were added. The algorithm converged after three iterations, yielding a network of 203 proteins and 350 interactions, selected from a set of 250 potential interactions. The resulting network, which we call the expanded network, is illustrated in Fig. 4A, showing the newly predicted proteins in red.

E

241 242

O

F

Fig. 3. Evaluating the accuracy of predictions. (A) Precision at top k predictions computed over held-out interactions for global and local classifiers. (B) Precision at top k for three sets of features: Phylogenic profiles, expression and protein signatures combining all three features improves precision for most values of k.

Fig. 4. Predicted interactions in the PSD. (A) The expanded network, containing proteins from the seed network (blue circles) and their interactions, and predicted interactions from a set of candidate interactions from public repositories (proteins marked in red circles). (B) The final network, containing the expanded network from (A) (blue circles) and also the top 30 predicted interactions between the expanded network and candidate proteins (in red). Proteins which have support for direct interactions in literature are marked by red circles, proteins connected by indirect interactions in red triangles and interactions with no previous supporting evidence are marked by red rectangles. (C) The neighborhood of Dlg4 from panel B. (D) The neighborhood of Src from panel B. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324

5

U

N C O

R

R

E

C

T

E

D

P

R O

O

F

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

6 t1:1 t1:2

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

Table 1 Supporting finding in literature for the top 30 most likely predictions.

t1:3

Protein 1

t1:4

Gene Protein name symbol

UniProt

t1:5

Grb2

t1:6

Dlg4

Growth factor receptor bound protein 2 PSD-95

t1:7

Actg1

t1:8

Grb2

t1:9

Grb2

t1:10

Grb2

t1:11

Mapk1

t1:12

Mapk1

t1:13

Protein 2 Reference

Calibrated score

Grabs (1997), Papin and Subramaniam (2004) Takeuchi et al. (1997), Naisbitt et al. (2000) Shen et al. (1998)

0.54

Q60631 Sntb2

Calcium/calmodulin dependent protein kinase II delta Syntrophin, basic 2

GRB2 interacts with Dnm1,Dnm2 Q9D0M5 PSD-95 interacts with SAPAP4 which interacts with DLC2 Q6PHZ2 Actin interacts with Camk2a Q61235 -

Q60631 Pcbp1

Poly(rC) binding protein 1

P60335

Suzuki et al. (2001)

Q60631 Pcbp2

Poly(rC) binding protein 2

Q61990

Mitogen-activated protein kinase 1 Mitogen-activated protein kinase 1

P63085

Cdk5

Cyclin-dependent kinase 5

P49615

P63085

Taok1

TAO kinase 1

Q5F2E8

Mapk1

Mitogen-activated protein kinase 1

P63085

Taok2

TAO kinase 2

t1:14

Dlg4

PSD-95

Q62108 Lrrtm4

Leucine-rich repeat transmembrane neuronal protein

t1:15

Dlg4

PSD-95

Q62108 Omg

t1:16

Dlg4

PSD-95

Q62108 Lrrtm1

Oligodendrocyte myelin glycoprotein Leucine-rich repeat transmembrane neuronal protein

t1:17 t1:18

Dlg4 Dnm1

PSD-95 Dynamin I

Q62108 Nlgn1 P39053 Prkca

Neuroligin-1 Protein kinase C, alpha

Q99K10 Q4VA93

t1:19

Grb2

Q60631 Tuba4a

Tubulin al4A chain

P68368

t1:20

Grb2

Q60631 Tuba1b

Tubulin, alpha 1B

P05213

–

0.402

t1:21

Grb2

Q60631 Kcnab1

–

0.4007

Src

Voltage-gated potassium channel subunit beta-1 Capping protein

P63143

t1:22

P24452

–

0.3774

t1:23

Dlg4

Growth factor receptor bound protein 2 Growth factor receptor bound protein 2 Growth factor receptor bound protein 2 Neuronal proto-oncogene tyrosine-protein kinase Src PSD-95

PSD-95 interacts with Leucine-rich repeat transmembrane neuronal protein2 PSD-95 interacts with Nlgn1 Dynamin interacts with Protein kinase C –

Q6PHZ2

Ewing et al. (2007)

0.3581

t1:24

Actg1

Actin, gamma, cytoplasmic 1 P63260

Flna

Dlg3/SAP102 interacts with CaM kinase II subunit alpha Filamin interacts with Actin

Davies et al. (1978)

0.3574

t1:25

Actg1

Actin, gamma, cytoplasmic 1 P63260

Rhog

0.3574

t1:26 t1:27

Actg1 Actg1

Actin, gamma, cytoplasmic 1 P63260 Actin, gamma, cytoplasmic 1 P63260

t1:28

Actg1

Actin, gamma, cytoplasmic 1 P63260

Bishop and Hall, (2000), Westphal et al. (2000) Ogata (1999) Rodríguez Del Castillo et al. (1992) Méjean et al. (1989)

t1:29 t1:30 t1:31 t1:32

Actg1 Actg1 Actg1 Actg1

Actin, gamma, cytoplasmic 1 Actin, gamma, cytoplasmic 1 Actin, gamma, cytoplasmic 1 Actin, gamma, cytoplasmic 1

t1:33

Src

t1:34

Src

Neuronal proto-oncogene tyrosine-protein kinase Src Neuronal proto-oncogene tyrosine-protein kinase Src

325 326 327 328 329

Q9UQ16

Q62108 Dynll2

Dynein light chain 2 (DLC2)

Actin, cytoplasmic 2

P63260

Growth factor receptor bound protein 2 Growth factor receptor bound protein 2 Growth factor receptor bound protein 2

O

Q62108 Camk2d

D

C

T

E

Q8K377

Calcium/calmodulin dependent protein kinase II delta Filamin-A; Actin-binding protein 280 Rho-related GTP-binding protein RhoG Actinin alpha 3 Actin-capping protein

Q8BTM8

GRB2 interacts with Poly binding protein1 GRB2 Indirectly interact with Poly binding protein 2 via PI3K regulatory subunit – TAO kinase 1 interacts with Mek3, upstream in the Map pathway (Tao1-Mek3-Mek2-Map) TAO kinase 2 interacts with Mek3 upstream in the Map pathway (Tao2-Mek3-Mek2-Map) PSD-95 interacts with Leucine-rich repeat transmembrane neuronal protein –

P05064

P63260 P63260 P63260 P63260

Fructose-bisphosphate aldolase A Aldoart1 Aldolase 1 A retrogene 1 Aldoart2 Aldolase 1 A retrogene 2 Rap1a Prkca Protein kinase C alpha type

P05480

Pcbp1

Poly(rC) binding protein 1

P60335

Indirect — Rac, Wave-1 (Wasf1), Actin Actin interacts with Actinin Direct via homologous — Scinderin interacts with Actin Actin interacts with Aldolase – – – Actin interacts with Protein kinase C –

P05480

Pcbp2

Poly(rC) binding protein 2

Q61990

–

C

N

U

Q63912

E

R

Capg

Q80XG9

R

Camk2d

Actn3 Capg Aldoa

interacting, for which we found no evidence for a direct interaction, but we did find evidence of indirect interactions through a third protein. Such association was found for example between PSD-95/Dlg4 and Dynein light chain 2/Dynll2. PSD-95 interacts with SAPAPs (Takeuchi et al., 1997), which interact with Dynein light chain (Naisbitt et al., 2000).

P84096 Q9JI91 P24452

P20444

0.4856 0.4692 0.4653

F

Dynamin-3

Bandyopadhyay et al. (2010), Papin and Subramaniam (2004)

O

Q60631 Dnm3

Interaction in literature

R O

UniProt

P

Protein name

P05480

Gene symbol

0.4653 0.4653

0.4403

Hutchison, (1998)

0.4403

Chen and Cobb (2001)

0.4403

Arbuckle et al. (2010)

0.4221

0.4221 De Wit et al. (2009)

0.4221

Irie (1997)

0.4215

0.402

Baatout (et al.)

0.3554 0.3547 0.3547 0.3547 0.3547 0.3529 0.3417 0.3155 0.3155

Discussion

330

We described a computational framework to reconstruct a PPI net- 331 work layer by layer starting from a high confidence “seed” network, 332 and used this framework to predict new proteins and interactions of 333

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

7

R O

O

F

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

E

T

C E R

339 340

involved in glutamate receptors trafficking and modulation (Sager et al., 2009; Lin et al., 2006). LRRTMs are leucine-rich repeat transmembrane neuronal proteins, predominantly expressed in mammals central nervous system (Haines and Rigby, 2007; Laurén et al., 2003; Linhoff et al., 2009). The predicted interaction between Lrrtm1 and PSD-95 is based on several sources of evidence. First, Lrrtm1 shares similar sequence with other proteins of the LRRTM family that were

R

337 338

the mouse PSD, yielding a 35% increase in the number of proteins over the initial high confidence “seed” network. We used the expanded network to infer possible interactions with candidate proteins identified in the PSD, and found experimental evidence for 18 of the top 30 suggested interactions. A particularly interesting inferred interaction was found between PSD-95 and Lrrtm1 (Fig. 4C). PSD-95 is a major scaffolding molecule

N C O

335 336

U

334

D

P

Fig. 5. Evaluating feature predictive power and prediction accuracy. (A) The number of phenotypes in common in pairs of interacting proteins (blue), and in pairs of non-interacting proteins (red) in the seed network. The bars corresponding to zero common phenotypes are omitted for clarity. (B) As in panel A, but for the expanded network. (C) Gene set enrichment analysis (GSEA) for phenotypes (see Experimental methods). (D) The distribution of linear correlation between gene expression vectors of protein pairs in the seed network: blue — interacting pairs, green — non interacting pairs. (E) Distribution of correlations as in panel D, but for the expanded network. (F) GSEA for gene expression in brain-related experiments (see Experimental methods). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 6. Evaluating predicted interactions. (A–B) Characteristics of a pair that is predicted to interact: Dlg4 (blue) and Lrrtm4 (green). (A) Expression profiles of brain related experiments. Data was scaled (zero mean, unit variance) for better visualization. (B) Phylogenetic profiles from (O'Brien et al, 2005). (C–E) Evaluating the prediction accuracy in expanded and seed networks and matching it to the calibrated score obtained by the trained models. (C) The precision at top k ranked interaction in the seed network. Precision is above 0.1 for k > 370. (D) Calibrated scores (see Experimental methods) in the seed network. For k = 370, the probability of interaction is estimated to be about 8%. (E) The precision at top k as compared to candidate interactions in the expanded network (see Experimental methods). Precision is above 0.3 for k = 30. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

341 342 343 344 345 346 347

371 372 373

Q11374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404

Q12405 406 407 408 409 410 411 412 413

F

O

R O

369 370

Training classifiers

414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 Q13 430

We examined two approaches based on the support vector machine (SVM) binary classifier. First, Global SVM trains a single interaction prediction model for the whole network. The input for the algorithm is a graph G(P,E), where nodes are proteins and edges are interactions, a set of proteins {p1,…pn} each represented by feature vectors pi∈ R d, a set of positive examples corresponding to pairs of proteins that interact, (pi, pj)∈ E, and negative examples (pk, pm)∉E. Second, Local SVM is an alternative approach that trains an individual model per node (Bleakley et al., 2007). For each protein pi∈ P, we train a model Mi, where positive examples correspond to proteins that interact with pi, (pi, pj)∈ E, and negative examples do not interact with it (pi, pk)∉ E. To compare between scores of SVM models trained for different nodes, we calibrated the scores of each SVM model to the range [0,1] using Platt's algorithm (Platt, 1999). The global model has the advantage that it has much more data to train on, and therefore may learn more robust models, and the local approach has the advantage of being more flexible to assign different weights to features at different parts of the protein network. Indeed, different feature types were found to be more discriminating for different interactions (see Supplementary Fig. 2). To compensate for the unbalanced number of positive and negative samples, we assigned different costs C + and C − for false positive and false negative types of mistakes. (Vapnik, 1998). Specifically, we tuned the value of a hyper parameter C over the grid {0.1, 1, 10,100} and set the value of C + to be C times the fraction of negatives, and C − to C times the fraction of positives. We also tuned the parameter γ corresponding to the width of the RBF kernel in the range {0.01, 0.1, 1, 5}. We evaluate the classification quality using two levels of 5-fold cross validation, one to tune the hyper parameters C and γ and one to tune the parameters of the classifiers.

431

Collecting and processing features for training

462

P

367 368

Experimental methods

D

365 366

of weights is fit to the whole network. In many studies, feature weights are analyzed to discover a small set of highly predictive features, but the above results suggest that such analysis may be misleading, since different parts of the network may have very different predictive features. The reconstruction framework described above is not limited to the mouse PSD, but can be extended to other organisms and organelles, including protein networks on the presynaptic side (Abul-Husn et al., 2009). Furthermore, reconstructed networks focus mostly on static interactions, mapping the set of potential interactions among PSD proteins. In reality, some interactions are transient and may only occur under specific conditions. The models we trained take into account correlations in features that are condition dependent, like transcription profiles across conditions or across brain regions. These models can therefore be further refined to a learned model that predicts conditionspecific interactions.

T

363

Q10364

C

361 362

E

359 360

R

357

Q9 358

R

355 356

O

354

C

352 353

N

350 351

found to bind PSD-95. Specifically, Lrrtm2 contains a similar domain structure (Lauren et al., 2003) and was shown to bind PSD-95 (De Wit et al., 2009). When comparing the protein sequences of mouse Lrrtm1 and Lrrtm2 from UniProt (Magrane and Consortium, 2011), we found that their similarity is highly significant (68%, BLAST (Altschul et al., 1990)), and that the specific domain in Lrrtm2 that binds PSD-95 (ECEV) is preserved in Lrrtm1. A second and independent source of evidence supporting the interaction between Lrrtm1 and PSD-95 is that the correlation between Lrrtm1 and Lrrtm2 Phylogenic profiles is very high (Pearson rho = 0.7818). LRRTMs are of special interest due to their possible involvement in psychiatric disorders. In particular, Lrrtm1, an imprinted gene on chromosome 2p12, was linked to Schizophrenia (SZ) by genetic association mapping (Francks et al., 2007) and also in a set of experiments examining Lrrtm1 knockout mice (Linhoff et al., 2009). While for many years the prevailing hypothesis of schizophrenia involved dopamine hyperactivity (Seeman, 1987), an alternative explanation involving NMDA receptor hypofunction emerged following pharmacological studies in which healthy subjects were exposed to NMDAR antagonists. In separate experiments, NMDAR antagonists Ketamine and phencyclidine (PSP) had been shown to induce schizophrenia-like symptoms (Javitt and Zukin, 1991; Krystal et al., 1994). Further studies, including postmortem (Clinton and Meador-Woodruff, 2004), neuroimaging (Pilowsky et al., 2006) and clinical studies have strengthened the hypothesis that NMDA is involved in schizophrenia. In search of the mechanism behind NMDAR hypoactivity, Chang-Gyu Hahn et al. found reduced tyrosine phosphorylation of NMDAR subunit following an increase in ErbB4 activation induced by Neuregulin1 (Hahn et al., 2006). Pitcher et al. (2011) have found NRG1-ErbB4 signaling to inhibit Src, a major NMDAR regulator (Yu et al., 2004), thus suppresses the enhancement of NMDA receptors activity. The inferred Lrrtm1-PSD-95 interaction and the association between Lrrtm1 and schizophrenia (Francks et al., 2007) suggest an alternative explanation for NMDAR down regulation in schizophrenia. Supplemental Fig. S1 illustrates the relation between these proteins. Under this hypothesis, a decrease in NMDAR tyrosine phosphorylation is mediated by Lrrtm1, PSD-95 and Src. Furthermore, the fact that Lrrtm1 can bind certain isoforms of presynaptic Neurexins (Siddiqui et al., 2010) may suggest a trans-synaptic NMDAR regulation mechanism mediated by Neurexin/Lrrtm1, which resembles the Neurexin/Neuroligin junction proposed by Südhof (2008). The difference in expression pattern of LRRTM family members across brain structures (Laurén et al., 2003), may imply that this type of regulation mechanism is mediated by a different member of the LRRTM family in different brain regions. A second example of an interesting inferred interaction, involves Src, and Capg (Fig. 4D), a protein that caps actin filaments, suggesting an alternative manner by which Src may be involved in control of dendritic spine shape. Dendritic spines are the primary location of excitatory synapse (Bourne and Harris, 2008). They start as a thin and transient filopodia (microspines) and change their shape in a way that is correlated with synaptic activity (Kasai et al., 2003). Actin regulation plays a pivotal role in this transformation (Hotulainen and Hoogenraad, 2010), particularly by controlling the actin filament length, since the formation of a mushroom-shaped spine head requires short actins. Actin capping protein (Capg) was shown to inhibit elongation of actin filaments by binding to the barbed ends of the filaments and keeping their filaments short (Fan et al., 2011). Src was previously found to regulate actin filaments via tyrosine phosphorylation of Cortactin, a filamentous actin binding protein (Huang et al., 1997). The new interaction predicted here between Src and Capg suggests an additional mechanism by which Src may control spine maturation. The computational approach taken in this paper is based on learning a separate model for each given protein in the network. This rich set of models allows weighting different features for different proteins and it significantly outperforms an approach where one common set

U

348

Q8 349

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

E

8

432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 Q14 451 452 453 454 455 456 457 458 459 460 461

To construct a feature vector for each protein, we collected datasets 463 from three sources: 464 (1) Gene expression data was downloaded from NCBI Geo (Edgar, 2002). We retrieved microarray expression profiles that hold the results of 471 brain related experiments conducted using the GPL1261 platform (Mus musculus Affymetrix Mouse Genome 430 2.0). For each experiment the expression vector was normalized using Box–Cox transformation (Box and Cox, 1964) and scaled using zero mean and unit variance. Missing values were imputed by filling the values of the closest column in Euclidean distance (Hastie et al., 1999).

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

465 466 467 468 469 470 471 472 473

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

485 486 487 Q15 488 489 490 491 492 493 494 495 496 497 498 499

511

Features used for evaluation

512

528

We collected data from three additional sources: phenotype annotations were downloaded from MGI (Blake et al., 2009) on Apr. 5, 2011. We used the leaf terms of the following brain related major categories: behavior/neurological phenotype (5386), nervous system phenotype (3631),hearing/vestibular/ear phenotype (5377), taste/ olfaction phenotype (5394), vision/eye phenotype (5391), craniofacial phenotype (5382), and embryogenesis phenotype (5380). Overall, 1735 terms were considered. Expression levels across brain structures were retrieved on-line from the Allen brain atlas (Lein et al., 2007). Additional gene expression profiles were downloaded from NCBI Geo (Barrett et al., 2005), containing 459 microarray expression profiles that describe the results of brain related experiments, as measured by the GPL81 NCBI platform. These experiments do not overlap with the ones used for model training and interaction prediction. The expression data for each experiment results column was normalized using Box– Cox transformation (Box and Cox, 1964) and scaled using zero mean and unit variance.

529

Selecting candidate interactions from large scale repositories

530

We retrieved protein interactions in the mouse from two databases: IntAct (Aranda et al., 2010), and MINT (Chatr-aryamontri et al., 2007), and used the MINT scoring system (Braun et al., 2009) to exclude interactions with scores below 0.6. We used the IntAct annotations to exclude interactions that were expanded from co-complexes

513 514 515 516 517 518 519 520 521 522 523 524 525 526 527

Q16 531 Q17 532 533 534

C

E

508 509

R

506 507

R

504 505

N C O

502 503

U

500 501

T

510

For each dataset we constructed a matrix of size NxFi, where N is the number of proteins in the network, and Fi is the number of features in a given dataset. Overall, we built three matrices, MExpr, MInterPro, and MPhylo, with the same number of rows and different number of columns. These matrices were combined into a single feature matrix in the following way. For LocalSVM, each dataset was transformed into a positive semi-definite matrix of similarities between genes, as described in Bleakley et al. (2007). All data was integrated by summing all matrices. KInt = KExpr + KInterPro + KPhylo. For Global SVM, all datasets were first concatenated into a matrix of size NΣi·Fi, each row in the matrix is a feature representation of a single node. Then, the vectors of all possible pairs were concatenated, yielding a matrix of size 2N(N-1)Σi Fi.

535 536 537 538 539 540 541 542 543 544 545

Potential interactions between the network and experimentally identified 546 PSD proteins 547

F

483 484

We collected proteins from three pull-down experiments. 118 proteins purified from mouse forebrain using tandem affinity purification (Fernández et al., 2009), 276 proteins identified in rat brain using single step immune-precipitation (Dosemeci et al., 2007). And 748 consensual hPSD proteins isolated from postsynaptic density in human neocortex (Bayés et al., 2010). Human proteins were converted to mouse homologues using biomaRt (Smedley et al., 2009), rat gene symbols required no translation as they share a unified nomenclature with mice (http://rgd.mcw.edu/nomen/nomen. shtml). After excluding proteins that lacked one or more features, or that were already present in the network, 765 proteins were left, yielding a total of (203 × 765) = 155,295 potential interactions.

548 549

Estimating the precision of predicted interactions

560

O

481 482

R O

480

P

478 479

n-ary interactions. We also excluded duplicated interactions, interactions containing proteins that are not expressed in brain (according to Lein et al. (2007)), pairs in which neither protein appeared in the network and interactions containing proteins that completely lacked data for one or more features. To decide which of the interactions will be added to the network, we set a calibrated score threshold. To estimate the lowest calibrated score for which precision at top k is still above 0.1, we addressed the seed network. We located the last point before precision at top k deteriorates below 0.1 (k = 370, Fig. 6C), and found the calibrated score at this point (score = 0.0815, Fig. 6D).

D

476 477

(2) A file containing 21,178 protein domains and family assignments (“signatures”) was downloaded from the InterPro database (Apweiler, 2001) on January 24th 2011. We represented each protein as a vector of weighted signatures using TF-IDF weighting, a procedure borrowed from information retrieval (Salton and Mcgill, 1986) and text mining (Dumais et al., 1998). In our context, the signatures serve as “terms”, proteins as “documents” and the entire dataset as “corpus”. Thus — TFij = dij / Σk(dkj), where dij is the number of occurrences of signature di in a protein pj, and Σk (dkj) is the number of signatures in protein pj. IDFj = log(|P| / |{pj: di∈ pj}|) where |P| is the total number of proteins in the dataset, and |{pj: di∈ pj}| is the number of proteins in which the signature di appears. The TF-IDF weight of a signature i in a protein j is therefore (TF-IDF)ij = TFij / IDFj. (3) Pairwise ortholog maps of 99 species were downloaded on December 1st 2010 from the Inparanoid database (O'Brien et al., 2005). For each gene, we calculated the orthology score by multiplying the gene's confidence score by the confidence level of its paralog cluster (ortholog group bootstrap value). In the case where an ortholog does not exist, the score was set to zero. We created a Phylogenic table of all mouse genes, as given in MGI (Blake et al., 2011), and their ortho-scores against all other 98 orthologs.

E

474 475

9

550 551 552 553 554 555 556 557 558 559

To estimate the precision of predicted interactions, we treated the list of candidate interactions as ground truth, and compared these interactions to the top ranked predicted interactions between proteins in the seed network and proteins in the list of candidate interactions that do not appear in the seed network. Fig. 6E depicts the ratio of pairs known to interact out of the top k predicted interactions. More than a third of the top 30 ranked pairs match interactions in the set of candidate interactions.

561 562

Network visualization

569

The full network of protein interactions is dense and therefore hard to visualize. To provide a more clear visualization of the close neighborhood of Dlg4 and Src (Fig. 4C, D), we applied GLay (Su et al., 2010), a structure clustering Cytoscape plugin. GLay decomposes the network into “communities” of densely interacting nodes. Applying the algorithm to the final inferred network (Fig. 4B), produced 11 clusters, which contain all the original network nodes, and 76% of the 380 edges. The full set of interactions and a high resolution version of the network visualization are available at http://chechiklab.biu.ac. il/~ossnat/psd_reconstruction.html.

570 571

Computational assessment of predicted interactions

580

Gene set enrichment analysis (GSEA) tests if a given set of gene S is enriched at the top of a ranked set L (Subramanian et al., 2005). Here L is the set of all possible pairs of network and candidate proteins ranked by their calibrated scores, and S is the subset of pairs that are highly similar based on an experimental assay (feature). Specifically, for gene expression, S was the set of all pairs for which the Pearson correlation between their gene expression profiles was above 0.3. For phenotypes, S contains all pairs that are involved in creating at least one common phenotype. Following Subramanian et al., we empirically estimated the statistical significance of the GSEA score using 1000 permutations using an exponent P = 1.

581 582

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

563 564 565 566 567 568

572 573 574 575 576 577 578 579 Q18

583 584 585 586 587 588 589 590 591

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

The authors declare that they have no conflict of interest.

596

Acknowledgments

597 598 599

This work was supported by the Israeli Science Foundation Grants 1001/08 and 1090/12, and by a Marie Curie Reintegration Grant PIRG06-GA-2009-256566.

600

References

N

C

O

R

R

E

C

T

Abul-Husn, Noura, S., Bushlin, Ittai, Morón, José A., Jenkins, Sherry L., Dolios, Georgia, Wang, Rong, Iyengar, Ravi, Avi, Ma'ayan, Devi, Lakshmi A., 2009. Systems approach to explore components and interactions in the presynapse. Proteomics 9 (12), 3303–3315. http://dx.doi.org/10.1002/pmic.200800767 (June). Ali, Declan, W., Salter, Michael W., 2001. NMDA receptor regulation by Src kinase signalling in excitatory synaptic transmission and plasticity. Curr. Opin. Neurobiol. 11 (3), 336–342 (June). Altschul, S.F., Gish, W., Miller, W., Myers, E.W., Lipman, D.J., 1990. Basic local alignment search tool. J. Mol. Biol. 215 (3), 403–410 (October 5). Apweiler, R., 2001. The interpro database, an integrated documentation resource for protein families, domains and functional sites. Nucleic Acids Res. 29 (1), 37–40. http://dx.doi.org/10.1093/nar/29.1.37 (January 1). Arbuckle, Margaret, I., Komiyama, Noboru H., Delaney, Ada, Coba, Marcelo, Garry, Emer M., Rosie, Roberta, Allchorne, Andrew J., et al., 2010. The SH3 domain of postsynaptic density 95 mediates inflammatory pain through phosphatidylinositol-3-kinase recruitment. EMBO Rep. 11 (6), 473–478. http://dx.doi.org/10.1038/embor.2010.63 (June 1). Archakov, Alexander, I., Govorun, Vadim M., Dubanov, Alexander V., Ivanov, Yuri D., Veselovsky, Alexander V., Lewi, Paul, Janssen, Paul, 2003. Protein–protein interactions as a target for drugs in proteomics. Proteomics 3 (4), 380–391 (April). Baatout, S, B Chatelain, P Staquet, M Symann, and C Chatelain. “Interaction Between Protein Kinase C and Actin in Megakaryocyte Polyploidization.” Anticancer Research 19 (5B): 4193–8. Bader, Joel, S., Chaudhuri, Amitabha, Rothberg, Jonathan M., Chant, John, 2004. Gaining confidence in high-throughput protein interaction networks. Nat. Biotechnol. 22 (1), 78–85. http://dx.doi.org/10.1038/nbt924 (January). Bandyopadhyay, Sourav, Chiang, Chih-yuan, Srivastava, Jyoti, Gersten, Merril, White, Suhaila, Bell, Russell, Kurschner, Cornelia, et al., 2010. A human MAP kinase interactome. Nat. Methods 7 (10), 801–805. http://dx.doi.org/10.1038/nmeth.1506 (September 19). Barrett, Tanya, Suzek, Tugba O., Troup, Dennis B., Wilhite, Stephen E., Ngau, Wing-Chi, Ledoux, Pierre, Rudnev, Dmitry, Lash, Alex E., Fujibuchi, Wataru, Edgar, Ron, 2005. NCBI GEO: mining millions of expression profiles—database and tools. Nucleic Acids Res. 33 (Database issue), D562–D566. http://dx.doi.org/10.1093/nar/gki022 (January 1). Bayés, Àlex, van de Lagemaat, Louie N., Collins, Mark O., Croning, Mike D.R., Whittle, Ian R., Choudhary, Jyoti S., Grant, Seth G.N., 2010. Characterization of the proteome, diseases and evolution of the human postsynaptic density. Nat. Neurosci. 14 (1), 19–21. http://dx.doi.org/10.1038/nn.2719 (December 19). Ben-Hur, Asa, Stafford Noble, William, 2005. Kernel methods for predicting protein– protein interactions. Bioinforma.(Oxford, England) 21 (Suppl. 1), i38–i46. http:// dx.doi.org/10.1093/bioinformatics/bti1016 ((suppl_1) (June)). Bhardwaj, Nitin, Lu, Hui, 2005. Correlation between gene expression profiles and protein– protein interactions within and across genomes. Bioinforma. (Oxford, England) 21 (11), 2730–2738. http://dx.doi.org/10.1093/bioinformatics/bti398 (June 1). Bishop, A., Hall, A., 2000. Rho GTPases and Their Effector Proteins (June). Blake, Judith A., Bult, Carol J., Eppig, Janan T., Kadin, James A., Richardson, Joel E., 2009. The mouse genome database genotypes: phenotypes. Nucleic Acids Res. 37 (Database issue), D712–D719. http://dx.doi.org/10.1093/nar/gkn886 (January). Blake, Judith A., Bult, Carol J., Kadin, James A., Richardson, Joel E., Eppig, Janan T., 2011. The Mouse Genome Database (MGD): premier model organism resource for mammalian genomics and genetics. Nucleic Acids Res. 39 (Database issue), D842–D848. http://dx.doi.org/10.1093/nar/gkq1008 (January). Bleakley, Kevin, Biau, Gérard, Vert, Jean-Philippe, 2007. Supervised reconstruction of biological networks with local models. Bioinforma. (Oxford, England) 23 (13), i57–i65. http://dx.doi.org/10.1093/bioinformatics/btm204 (July 1). Bourne, Jennifer, N., Harris, Kristen M., 2008. Balancing structure and function at hippocampal dendritic spines. Annu. Rev. Neurosci. 31, 47–67 (January 17). Box, G.E.P., Cox, D.R., 1964. An analysis of transformations. J. R. Stat. Soc. Ser. B Stat. Methodol. 26, 211–252 (October 19). Chatr-aryamontri, Andrew, Ceol, Arnaud, Montecchi Palazzi, Luisa, Nardelli, Giuliano, Victoria Schneider, Maria, Castagnoli, Luisa, Cesareni, Gianni, 2007. MINT: the Molecular INTeraction database. Nucleic Acids Res. 35 (Database issue), D572–D574. http://dx.doi.org/10.1093/nar/gkl950 (January). Chen, Z., Cobb, M.H., 2001. Regulation of stress-responsive mitogen-activated protein (MAP) kinase pathways by TAO2. J. Biol. Chem. 276 (19), 16070–16075. http://dx.doi.org/10.1074/jbc.M100681200 (May 11).

U

601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 Q19 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 Q20 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666

F

595

O

Conflict of interest

R O

594

Chen, Xue-Wen, Liu, Mei, 2005. Prediction of protein–protein interactions using random decision forest framework. Bioinforma. (Oxford, England) 21 (24), 4394–4400. http://dx.doi.org/10.1093/bioinformatics/bti721 (December 15). Cheng, Dongmei, Hoogenraad, Casper C., Rush, John, Ramm, Elizabeth, Schlager, Max A., Duong, Duc M., Xu, Ping, et al., 2006. Relative and absolute quantification of postsynaptic density proteome isolated from rat forebrain and cerebellum. Molecular & Cellular Proteomics: MCP 5 (6), 1158–1170. http://dx.doi.org/10.1074/ mcp.D500009-MCP200 (July). Cho, Chang-Hoon, St-Gelais, Fannie, Zhang, Wei, Tomita, Susumu, Howe, James R., 2007. Two families of TARP isoforms That have distinct effects on the kinetic properties of AMPA receptors and synaptic currents. Neuron 55 (6), 890–904. http://dx.doi.org/10.1016/j.neuron.2007.08.024 (September 20). Clinton, Sarah M., Meador-Woodruff, James H., 2004. Abnormalities of the NMDA Receptor and associated intracellular molecules in the thalamus in schizophrenia and bipolar disorder. Neuropsychopharmacology 29 (7), 1353–1362 (July 31). Collins, Mark O., Husi, Holger, Yu, Lu, Brandon, Julia M., Anderson, Chris N.G., Blackstock, Walter P., Choudhary, Jyoti S., Grant, Seth G.N., 2006. Molecular characterization and comparison of the components and multiprotein complexes in the postsynaptic proteome. J. Neurochem. 97 (Suppl. 1), 16–23. http://dx.doi.org/ 10.1111/j.1471 (April). Davies, P.J., Wallach, D., Willingham, M.C., Pastan, I., Yamaguchi, M., Robson, R.M., 1978. Filamin-actin interaction. Dissociation of binding from gelation by Ca2+-activated proteolysis. J. Biol. Chem. 253 (11), 4036–4042. De Wit, Joris, Sylwestrak, Emily, O'Sullivan, Matthew L., Otto, Stefanie, Tiglio, Katie, Savas, Jeffrey N., Yates, John R., Comoletti, Davide, Taylor, Palmer, Ghosh, Anirvan, 2009. LRRTM2 Interacts with neurexin1 and regulates excitatory synapse formation. Neuron 64 (6), 799–806. http://dx.doi.org/10.1016/j.neuron.2009.12.019 (December 24). Dong, H., O'Brien, R.J., Fung, E.T., Lanahan, A.A., Worley, P.F., Huganir, R.L., 1997. GRIP: a synaptic PDZ domain-containing protein that interacts with AMPA receptors. Nature 386 (6622), 279–284. http://dx.doi.org/10.1038/386279a0 (March 20). Dosemeci, Ayse, Makusky, Anthony J., Jankowska-Stephens, Ewa, Yang, Xiaoyu, Slotta, Douglas J., Markey, Sanford P., 2007. Composition of the synaptic PSD-95 complex. Molecular and Cellular Proteomics: MCP 6 (10), 1749–1760. http://dx.doi.org/ 10.1074/mcp.M700040-MCP200 (October). Dumais, Susan, Platt, John, Heckerman, David, Sahami, Mehran, 1998. “Inductive Learning Algorithms and Representations for Text Categorization”. In Proceedings of the Seventh International Conference on Information and Knowledge Management — CIKM. , 98. ACM Press, New York, New York, USA, pp. 148–155. http://dx.doi.org/ 10.1145/288627.288651. Edgar, R., 2002. Gene expression omnibus: NCBI Gene expression and hybridization array data repository. Nucleic Acids Res. 30 (1), 207–210. http://dx.doi.org/10.1093/ nar/30.1.207 (January 1). Ewing, Rob M., Chu, Peter, Elisma, Fred, Li, Hongyan, Taylor, Paul, Climie, Shane, McBroom-Cerajewski, Linda, et al., 2007. Large-scale mapping of human protein– protein interactions by mass spectrometry. Mol. Syst. Biol. 3, 89. http://dx.doi.org/ 10.1038/msb4100134 (January 13). Fan, Yanjie, Tang, Xin, Vitriol, Eric, Chen, Gong, Zheng, James Q., 2011. Actin capping protein is required for dendritic spine development and synapse formation. J. Neurosci. Off. J. Soc. Neurosci. 31 (28), 10228–10233 (July 13). Fernández, Esperanza, Collins, Mark O., Uren, Rachel T., Kopanitsa, Maksym V., Komiyama, Noboru H., Croning, Mike D.R., Zografos, Lysimachos, Douglas Armstrong, J., Choudhary, Jyoti S., Grant, Seth G.N., 2009. Targeted tandem affinity purification of PSD-95 recovers core postsynaptic complexes and schizophrenia susceptibility proteins. Mol. Syst. Biol. 5, 269. http://dx.doi.org/10.1038/msb.2009.27 (January 19). Francks, C., Maegawa, S., Laurén, J., Abrahams, B.S., Velayos-Baeza, A., Medland, S.E., Colella, S., et al., 2007. LRRTM1 on chromosome 2p12 Is a maternally suppressed gene that is associated paternally with handedness and schizophrenia. Molecular Psychiatry 12 (12), 1129–1139. http://dx.doi.org/10.1038/sj.mp.4002053 (December 31, 1057). Ge, H., Liu, Z., Church, G.M., Vidal, M., 2001. Correlation between transcriptome and interactome mapping data from Saccharomyces cerevisiae. Nat. Genet. 29 (4), 482–486. http://dx.doi.org/10.1038/ng776 (December 1). Goh, C.S., Bogan, A.A., Joachimiak, M., Walther, D., Cohen, F.E., 2000. Co-evolution of proteins with their interaction partners. J. Mol. Biol. 299 (2), 283–293. http://dx.doi.org/ 10.1006/jmbi.2000.3732 (June 2). Grabs, D., 1997. The SH3 domain of amphiphysin binds the proline-rich domain of dynamin at a single site that defines a new SH3 binding consensus sequence. J. Biol. Chem. 272 (20), 13419–13425. http://dx.doi.org/10.1074/jbc.272.20.13419 (May 16). Hahn, Chang-Gyu, Wang, Hoau-Yan, Cho, Dan-Sung, Talbot, Konrad, Gur, Raquel E., Berrettini, Wade H., Bakshi, Kalindi, et al., 2006. Altered neuregulin 1-erbB4 signaling contributes to NMDA receptor hypofunction in schizophrenia. Nat. Med. 12 (7), 824–828. http://dx.doi.org/10.1038/nm1418 (July). Haines, Bryan P., Rigby, Peter W.J., 2007. Developmentally regulated expression of the LRRTM gene family during mid-gestation mouse embryogenesis. Gene Expression Patterns: GEP 7 (1–2), 23–29 (January). Hastie, Trevor, Tibshirani, Robert, Sherlock, Gavin, Eisen, Michael, Brown, Patrick, Botstein, David, 1999. Imputing missing data for gene expression arrays. Hermjakob, Henning, Montecchi-Palazzi, Luisa, Lewington, Chris, Mudali, Sugath, Kerrien, Samuel, Orchard, Sandra, Vingron, Martin, et al., 2004. IntAct: an open source molecular interaction database. Nucleic Acids Res. 32 (Database issue), D452–D455. http://dx.doi.org/10.1093/nar/gkh052 (January 1). Hormozdiari, Fereydoun, Salari, Raheleh, Bafna, Vineet, Cenk Sahinalp, S., 2010. Protein–protein interaction network evaluation for identifying potential drug targets. J. Comput. Biol. 17 (5), 669–684 (May 25).

P

Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.mcn.2013.04.004.

D

592 593

E

10

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

N C O

R

R

E

C

D

P

R O

O

F

dependent NMDA receptor NR2B to NR2A subunit switch. Neuron 70 (2), 339–351 (April 28). Méjean, C., Pons, F., Benyamin, Y., Roustan, C., 1989. Antigenic probes locate binding sites for the glycolytic enzymes glyceraldehyde-3-phosphate dehydrogenase, aldolase and phosphofructokinase on the actin monomer in microfilaments. Biochem. J. 264 (3), 671–677 (December 15). Naisbitt, Scott, Valtschanoff, Juli, Allison, Daniel W., Sala, Carlo, Kim, Eunjoon, Craig, Ann Marie, Weinberg, Richard J., Sheng, Morgan, 2000. Interaction of the postsynaptic density-95/guanylate kinase domain-associated protein complex with a light chain of myosin-v and dynein. J. Neurosci. 20 (12), 4524–4534. Ng, S.-K., Zhang, Z., Tan, S.-H., 2003. Integrative approach for computationally inferring protein domain interactions. Bioinformatics 19 (8), 923–929. http://dx.doi.org/ 10.1093/bioinformatics/btg118 (May). Nishimune, Atsushi, Isaac, John T.R., Molnar, Elek, Noel, Jacques, Nash, S.Russell, Tagaya, Mitsuo, Collingridge, Graham L., Nakanishi, Shigetada, Henley, Jeremy M., 1998. NSF Binding to GluR2 Regulates Synaptic Transmission. Neuron 21 (1), 87–97. http://dx.doi.org/10.1016/S0896-6273(00)80517-6 (July 1). O'Brien, Kevin, P., Remm, Maido, Sonnhammer, Erik L.L., 2005. In paranoid: a comprehensive database of eukaryotic orthologs. Nucleic Acids Res. 33 (Database issue), D476–D480. http://dx.doi.org/10.1093/nar/gki107 (January 1). Ogata, H., 1999. KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 27 (1), 29–34. http://dx.doi.org/10.1093/nar/27.1.29 (January 1). Papin, Jason, Subramaniam, Shankar, 2004. Bioinformatics and cellular signaling. Curr. Opin. Biotechnol. 15 (1), 78–81. http://dx.doi.org/10.1016/j.copbio.2004.01.003 (February). Peng, Junmin, Jong Kim, Myung, Cheng, Dongmei, Duong, Duc M., Gygi, Steven P., Sheng, Morgan, 2004. Semiquantitative proteomic analysis of rat forebrain postsynaptic density fractions by mass spectrometry. J. Biol. Chem. 279 (20), 21003–21011. http://dx.doi.org/10.1074/jbc.M400103200 (May 14). Pilowsky, L.S., Bressan, R.A., Stone, J.M., Erlandsson, K., Mulligan, R.S., Krystal, J.H., Ell, P.J., 2006. First in vivo evidence of an NMDA receptor deficit in medication-free schizophrenic patients. Mol. Psychiatry 11 (2), 118–119 (February 27). Pitcher, Graham M., Kalia, Lorraine V., Ng, David, Goodfellow, Nathalie M., Yee, Kathleen T., Lambe, Evelyn K., Salter, Michael W., 2011. Schizophrenia susceptibility pathway neuregulin 1-ErbB4 suppresses Src upregulation of NMDA receptors. Nat. Med. 17 (4), 470–478. http://dx.doi.org/10.1038/nm.2315 (April 27). Platt, John C., 1999. Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods. MIT Press. Pocklington, Andrew J., Cumiskey, Mark, Douglas Armstrong, J., Grant, Seth G.N., 2006. The proteomes of neurotransmitter receptor complexes form modular networks with distributed functionality underlying plasticity and behaviour. Mol. Syst. Biol. 2, 0023. http://dx.doi.org/10.1038/msb4100041 (January 2006). Qi, Yanjun, Bar-Joseph, Ziv, Klein-Seetharaman, Judith, 2006. Evaluation of different biological data and computational classification methods for use in protein interaction prediction. Proteins 63 (3), 490–500. http://dx.doi.org/10.1002/prot.20865 (May 15). Rodríguez Del Castillo, A., Vitale, M.L., Trifaró, J.M., 1992. Ca2+ and pH determine the interaction of chromaffin cell Scinderin with phosphatidylserine and phosphatidylinositol 4,5,-biphosphate and its cellular distribution during nicotinicreceptor stimulation and protein kinase C activation. J. Cell Biol. 119 (4), 797–810 (November). Sager, C., Tapken, D., Kott, S., Hollmann, M., 2009. Functional modulation of AMPA receptors by transmembrane AMPA receptor regulatory proteins. Neuroscience 158 (1), 45–54. http://dx.doi.org/10.1016/j.neuroscience.2007.12.046 (January 12). Saglietti, Laura, Dequidt, Caroline, Kamieniarz, Kinga, Rousset, Marie-Claude, Valnegri, Pamela, Thoumine, Olivier, Beretta, Francesca, et al., 2007. Extracellular interactions between GluR2 and N-cadherin in spine regulation. Neuron 54 (3), 461–477. http://dx.doi.org/10.1016/j.neuron.2007.04.012 (May 3). Sakarya, Onur, Armstrong, Kathryn A., Adamska, Maja, Adamski, Marcin, Wang, I-Fan, Tidor, Bruce, Degnan, Bernard M., Oakley, Todd H., Kosik, Kenneth S., 2007. A postsynaptic scaffold at the origin of the animal kingdom. In: Vosshall, Leslie (Ed.), PLoS One 2 (6), e506. http://dx.doi.org/10.1371/journal.pone.0000506 (January). Sala, Carlo, Piëch, Valentin, Wilson, Nathan R., Passafaro, Maria, Liu, Guosong, Sheng, Morgan, 2001. Regulation of dendritic spine morphology and synaptic function by Shank and Homer. Neuron 31 (1), 115–130 (July). Salton, G., Mcgill, M.J., 1986. Introduction to modern information retrieval. Sato, Shigeru, Omori, Yoshihiro, Katoh, Kimiko, Kondo, Mineo, Kanagawa, Motoi, Miyata, Kentaro, Funabiki, Kazuo, et al., 2008. Pikachurin, a dystroglycan ligand, is essential for photoreceptor ribbon synapse formation. Nat. Neurosci. 11 (8), 923–931. http://dx.doi.org/10.1038/nn.2160 (August). Schulz, Torsten W., Nakagawa, Terunaga, Licznerski, Pawel, Pawlak, Verena, Kolleker, Alexander, Rozov, Andrei, Kim, Jinhyun, et al., 2004. Actin/alpha-actinin-dependent transport of AMPA receptors in dendritic spines: role of the PDZ-LIM protein RIL. J. Neurosci. 24 (39), 8584–8594. http://dx.doi.org/10.1523/JNEUROSCI.2100-04.2004 (September 29). Schwenk, Jochen, Harmel, Nadine, Zolles, Gerd, Bildl, Wolfgang, Kulik, Akos, Heimrich, Bernd, Chisaka, Osamu, et al., 2009. Functional Proteomics identify cornichon proteins as auxiliary subunits of AMPA receptors. Science 323 (5919), 1313–1319. http://dx.doi.org/10.1126/science.1167852 (March 6). Segal, E., Wang, H., Koller, D., 2003. Discovering molecular pathways from protein interaction and gene expression data. Bioinformatics 19 (Suppl. 1), i264–i272 (July 10). Setou, Mitsutoshi, Seog, Dae-Hyung, Tanaka, Yosuke, Kanai, Yoshimitsu, Takei, Yosuke, Kawagishi, Masahiko, Hirokawa, Nobutaka, 2002. Glutamate-receptor-interacting protein GRIP1 directly steers kinesin to dendrites. Nature 417 (6884), 83–87. http://dx.doi.org/10.1038/nature743 (May 2).

E

T

Hotulainen, Pirta, Hoogenraad, Casper C., 2010. Actin in dendritic spines: connecting dynamics to function. J. Cell Biol. 189 (4), 619–629 (May 17). Huang, C., Ni, Y., Wang, T., Gao, Y., Haudenschild, C., Zhan, X., 1997. Down-regulation of the filamentous actin cross-linking activity of cortactin by Src-mediated tyrosine phosphorylation. J. Biol. Chem. 272 (21), 13911–13915 (May 23). Hunt, David L., Castillo, Pablo E., 2012. Synaptic plasticity of NMDA receptors: mechanisms and functional implications. Curr. Opin. Neurobiol. 22 (3), 496–508 (June). Hutchison, M., 1998. Isolation of TAO1, a protein kinase that activates MEKs in stressactivated protein kinase cascades. J. Biol. Chem. 273 (44), 28625–28632. http:// dx.doi.org/10.1074/jbc.273.44.28625 (October 30). Irie, M., 1997. Binding of neuroligins to PSD-95. Science 277 (5331), 1511–1515. http://dx.doi.org/10.1126/science.277.5331.1511 (September 5). Jackson, Alexander C., Nicoll, Roger A., 2009. Neuroscience: AMPA receptors get ‘pickled’. Nature 458 (7238), 585–586. http://dx.doi.org/10.1038/458585a (April 2). Jansen, Ronald, Greenbaum, Dov, Gerstein, Mark, 2002. Relating whole-genome expression data with protein–protein interactions. Genome Res. 12 (1), 37–46. http://dx.doi.org/10.1101/gr.205602 (January). Jansen, Ronald, Yu, Haiyuan, Greenbaum, Dov, Kluger, Yuval, Krogan, Nevan J., Chung, Sambath, Emili, Andrew, Snyder, Michael, Greenblatt, Jack F., Gerstein, Mark, 2003. A Bayesian networks approach for predicting protein–protein interactions from genomic data. Science 302 (5644), 449–453. http://dx.doi.org/10.1126/science.1087361 (October 17). Javitt, D.C., Zukin, S.R., 1991. Recent advances in the phencyclidine model of schizophrenia. Am. J. Psychiatry 148 (10), 1301–1308 (October). Kalia, Lorraine V., Kalia, Suneil K., Salter, Michael W., 2008. NMDA receptors in clinical neurology: excitatory times ahead. Lancet Neurol. 7 (8), 742–755 (August). Kasai, Haruo, Matsuzaki, Masanori, Noguchi, Jun, Yasumatsu, Nobuaki, Nakahara, Hiroyuki, 2003. Structure-stability-function relationships of dendritic Spines. Trends Neurosci. 26 (7), 360–368 (July). Kristiansen, Lars V., Huerta, Ibone, Beneyto, Monica, Meador-Woodruff, James H., 2007. NMDA receptors and schizophrenia. Curr. Opin. Pharmacol. 7 (1), 48–55. http://dx.doi.org/10.1016/j.coph.2006.08.013 (February). Krystal, J.H., Karper, L.P., Seibyl, J.P., Freeman, G.K., Delaney, R., Bremner, J.D., Heninger, G.R., Bowers, M.B., Charney, D.S., 1994. Subanesthetic effects of the noncompetitive NMDA antagonist, ketamine, in humans: psychotomimetic, perceptual, cognitive, and neuroendocrine responses. Arch. Gen. Psychiatry 51 (3), 199–214. http://dx.doi.org/10.1001/archpsyc.1994.03950030035004 (March 1). Kulangara, Karina, Kropf, Michel, Glauser, Liliane, Magnin, Sarah, Alberi, Stefano, Yersin, Alexandre, Hirling, Harald, 2007. Phosphorylation of glutamate receptor interacting protein 1 regulates surface expression of glutamate receptors. J. Biol. Chem. 282 (4), 2395–2404 (January 26). Lauren, J., Airaksinen, M.S., Saarma, M., Timmusk, T.T., 2003. A novel gene family encoding leucine-rich repeat transmembrane proteins differentially expressed in the nervous system$star. Genomics 81 (4), 411–421. http://dx.doi.org/10.1016/ S0888-7543(03)00030-2 (April). Laurén, Juha, Airaksinen, Matti S., Saarma, Mart, Timmusk, Tõnis, 2003. A novel gene family encoding leucine-rich repeat transmembrane proteins differentially expressed in the nervous system. Genomics 81 (4), 411–421 (April). Legrain, Pierre, Wojcik, Jérôme, Gauthier, Jean-Michel, 2001. Protein–protein interaction maps: a lead towards cellular functions. Trends. Genet. 17 (6), 346–352 (June). Lein, Ed S., Hawrylycz, Michael J., Ao, Nancy, Ayres, Mikael, Bensinger, Amy, Bernard, Amy, Boe, Andrew F., et al., 2007. Genome-wide atlas of gene expression in the adult mouse brain. Nature 445 (7124), 168–176. http://dx.doi.org/10.1038/ nature05453 (January 11). Leonard, A.S., Monika, A.D., Horne, M.C., Craig, C.G., Hell, J.W., 1998. SAP97 is associated with the alpha-amino-3-hydroxy-5-methylisoxazole-4-propionic acid receptor GluR1 subunit. J. Biol. Chem. 273 (31), 19518–19524. http://dx.doi.org/10.1074/ jbc.273.31.19518 (July 31). Leonoudakis, Dmitri, Conti, Lisa R., Anderson, Scott, Radeke, Carolyn M., McGuire, Leah M.M., Adams, Marvin E., Froehner, Stanley C., Yates, John R., Vandenberg, Carol A., 2004. Protein trafficking and anchoring complexes revealed by proteomic analysis of inward rectifier potassium channel (Kir2.x)-associated proteins. J. Biol. Chem. 279 (21), 22331–22346. http://dx.doi.org/10.1074/jbc.M400285200 (May 21). Li, Wan, Ka, Hornshaw, Martin P., Van Der Schors, Roel C., Watson, Rod, Tate, Stephen, Casetta, Bruno, Jimenez, Connie R., et al., 2004. Proteomics analysis of rat brain postsynaptic density. Implications of the diverse protein functional groups for the integration of synaptic physiology. J. Biol. Chem. 279 (2), 987–1002. http://dx.doi.org/10.1074/jbc.M303116200 (January 9). Lin, Ying, Jover-Mengual, Teresa, Wong, Judy, Bennett, Michael V.L., Suzanne Zukin, R., 2006. PSD-95 and PKC converge in regulating NMDA receptor trafficking and gating. Proc. Natl. Acad. Sci. U. S. A. 103 (52), 19902–19907. http://dx.doi.org/ 10.1073/pnas.0609924104 (December 26). Linhoff, Michael W., Laurén, Juha, Cassidy, Robert M., Dobie, Frederick A., Takahashi, Hideto, Nygaard, Haakon B., Airaksinen, Matti S., Strittmatter, Stephen M., Marie Craig, Ann, 2009. An unbiased expression screen for synaptogenic proteins identifies the LRRTM protein family as synaptic organizers. Neuron 61 (5), 734–749. http://dx.doi.org/10.1016/j.neuron.2009.01.017 (March 12). Liu, J.P., Powell, K.A., Sudhof, T.C., Robinson, P.J., 1994. Dynamin I is a Ca(2+)-sensitive phospholipid-binding protein with very high affinity for protein kinase C. J. Biol. Chem. 269 (33), 21043–21050. Magrane, Michele, Uniprot Consortium, 2011. UniProt knowledgebase: a hub of integrated protein data. Database (Oxford) (2011 (0) (January 9): bar009). Malenka, Robert C., Bear, Mark F., 2004. LTP and LTD: an embarrassment of riches. Neuron 44 (1), 5–21 (September 30). Matta, Jose A., Ashby, Michael C., Sanz-Clemente, Antonio, Roche, Katherine W., Isaac, John T.R., 2011. mGluR5 and NMDA receptors drive the experience- and activity-

U

753 754 755 756 757 758 759 760 761 762 763 764 Q21 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 Q22 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838

11

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924

D

P

R O

O

F

Terashima, Akira, Cotton, Lucy, Dev, Kumlesh K., Meyer, Guido, Zaman, Shahid, Duprat, Fabrice, Henley, Jeremy M., Collingridge, Graham L., Isaac, John T.R., 2004. Regulation of synaptic strength and AMPA receptor subunit composition by PICK1. J. Neurosci. 24 (23), 5381–5390. http://dx.doi.org/10.1523/JNEUROSCI.4378-03.2004 (June). Torres, R., Firestein, B.L., Dong, H., Staudinger, J., Olson, E.N., Huganir, R.L., Bredt, D.S., Gale, N.W., Yancopoulos, G.D., 1998. PDZ proteins bind, cluster, and synaptically colocalize with Eph receptors and their ephrin ligands. Neuron 21 (6), 1453–1463. http://dx.doi.org/10.1016/S0896-6273(00)80663-7 (December). Torres, Gonzalo E., Yao, Wei-Dong, Mohn, Amy R., Quan, Hui, Kim, Kyeong-Man, Levey, Allan I., Staudinger, Jeff, Caron, Marc G., 2001. Functional interaction between monoamine plasma membrane transporters and the synaptic PDZ domain-containing protein PICK1. Neuron 30 (1), 121–134. http://dx.doi.org/10.1016/S0896-6273(01) 00267-7 (April). Traynelis, Stephen F., Wollmuth, Lonnie P., McBain, Chris J., Menniti, Frank S., Vance, Katie M., Ogden, Kevin K., Hansen, Kasper B., Yuan, Hongjie, Myers, Scott J., Dingledine, Ray, 2010. Glutamate receptor ion channels: structure, regulation, and function. Pharmacol. Rev. 62 (3), 405–496 (September 1). Uchino, Shigeo, Wada, Hidenori, Honda, Shizuyo, Nakamura, Yasuko, Ondo, Yumiko, Uchiyama, Takayoshi, Tsutsumi, Moe, Suzuki, Eri, Hirasawa, Takae, Kohsaka, Shinichi, 2006. Direct interaction of post-synaptic density-95/Dlg/ZO-1 domaincontaining synaptic molecule Shank3 with GluR1 alpha-amino-3-hydroxy-5methyl-4-isoxazole Propionic acid receptor. J. Neurochem. 97 (4), 1203–1214. http://dx.doi.org/10.1111/j.1471-4159.2006.03831.x (May). Van der Bliek, Alexander, M., 1999. Functional diversity in the dynamin family. Trends Cell Biol. 9 (3), 96–102. http://dx.doi.org/10.1016/S0962-8924(98)01490-1 (March). Vapnik, Vladimir, Vapnik, Vladimir, 1998. Statistical learning theory, 1st ed. WileyInterscience. Volk, Lenora, Kim, Chong-Hyun, Takamiya, Kogo, Yu, Yilin, Huganir, Richard L., 2010. Developmental regulation of protein interacting with C kinase 1 (PICK1) function in hippocampal synaptic plasticity and learning. Proc. Natl. Acad. Sci. U. S. A. 107 (50), 21784–21789 (December 14). Von Engelhardt, Jakob, Mack, Volker, Sprengel, Rolf, Kavenstock, Netta, Wan Li, Ka, Stern-Bach, Yael, Smit, August B., Seeburg, Peter H., Monyer, Hannah, 2010. CKAMP44: a brain-specific protein attenuating short-term synaptic plasticity in the dentate gyrus. Science (New York, N.Y.) 327 (5972), 1518–1522. http://dx.doi.org/ 10.1126/science.1184178 (March 19). Wang, Chang-Yu, Chang, Kai, Petralia, Ronald S., Wang, Ya-Xian, Seabold, Gail K., Wenthold, Robert J., 2006. A novel family of adhesion-like molecules that interacts with the NMDA receptor. J. Neurosci. 26 (8), 2174–2183. http://dx.doi.org/ 10.1523/JNEUROSCI.3799-05.2006 (February 22). Westphal, R.S., Soderling, S.H., Alto, N.M., Langeberg, L.K., Scott, J.D., 2000. Scar/WAVE-1, a Wiskott–Aldrich syndrome protein, assembles an actin-associated multi-kinase scaffold. EMBO J. 19 (17), 4589–4600. http://dx.doi.org/10.1093/emboj/19.17.4589 (September 1). Xia, Jun, Zhang, Xiaoqun, Staudinger, Jeff, Huganir, Richard L., 1999. Clustering of AMPA receptors by the synaptic PDZ domain-containing protein PICK1. Neuron 22 (1), 179–187. http://dx.doi.org/10.1016/S0896-6273(00)80689-3 (January 1). Yoshimura, Yoshiyuki, Yamauchi, Yoshio, Shinkawa, Takashi, Taoka, Masato, Donai, Hitomi, Takahashi, Nobuhiro, Isobe, Toshiaki, Yamauchi, Takashi, 2003. Molecular constituents of the postsynaptic density fraction revealed by proteomic analysis using multidimensional liquid chromatography–tandem mass spectrometry. J. Neurochem. 88, 759–768. http://dx.doi.org/10.1046/j.1471-4159.2003.02136.x (November 28). Yu, Haiyuan, Luscombe, Nicholas M., Xin Lu, Hao, Zhu, Xiaowei, Xia, Yu, Han, Jing-Dong J., Bertin, Nicolas, Chung, Sambath, Vidal, Marc, Gerstein, Mark, 2004. Annotation transfer between genomes: protein–protein interologs and protein–DNA regulogs. Genome Res. 14 (6), 1107–1118. http://dx.doi.org/10.1101/gr.1774904 (June).

R

R

E

C

T

Sharan, Roded, Suthram, Silpa, Kelley, Ryan M., Kuhn, Tanja, McCuine, Scott, Uetz, Peter, Sittler, Taylor, Karp, Richard M., Ideker, Trey, 2005. Conserved patterns of protein interaction in multiple species. Proc. Natl. Acad. Sci. U. S. A. 102 (6), 1974–1979. http://dx.doi.org/10.1073/pnas.0409522102 (February 8). Shen, K., Teruel, M., Subramanian, K., Meyer, T., 1998. CaMKII functions as an F-Actin targeting module that localizes CaMKII heterooligomers to dendritic spines. Neuron 21 (3), 593–606. http://dx.doi.org/10.1016/S0896-6273(00)80569-3 (September 1). Sheng, Morgan, Jong Kim, Myung, 2002. Postsynaptic signaling and plasticity mechanisms. Science 298 (5594), 776–780 (October 25). Siddiqui, Tabrez J., Pancaroglu, Raika, Kang, Yunhee, Rooyakkers, Amanda, Marie Craig, Ann, 2010. LRRTMs and neuroligins bind neurexins with a differential code to cooperate in glutamate synapse development. J. Neurosci. 30 (22), 7495–7506. http://dx.doi.org/10.1523/JNEUROSCI.0470-10.2010 (June 2). Silverman, Joshua B., Restituito, Sophie, Lu, Wei, Lee-Edwards, Laveria, Khatri, Latika, Ziff, Edward B., 2007. Synaptic anchorage of AMPA receptors by cadherins through neural plakophilin-related arm protein AMPA receptor-binding protein complexes. J. Neurosci. 27 (32), 8505–8516. http://dx.doi.org/10.1523/JNEUROSCI.1395-07.2007 (August 8). Skrabanek, Lucy, Saini, Harpreet K., Bader, Gary D., Enright, Anton J., 2008. Computational prediction of protein–protein interactions. Mol. Biotechnol. 38 (1), 1–17. http://dx.doi.org/10.1007/s12033-007-0069-2 (January 1). Smedley, Damian, Haider, Syed, Ballester, Benoit, Holland, Richard, London, Darin, Thorisson, Gudmundur, Kasprzyk, Arek, 2009. BioMart—biological queries made easy. BMC Genomics 10 (1), 22. http://dx.doi.org/10.1186/1471-2164-10-22 (January). Song, Insuk, Kamboj, Sunjeev, Xia, Jun, Dong, Hualing, Liao, Dezhi, Huganir, Richard L., 1998a. Interaction of the N-ethylmaleimide–sensitive factor with AMPA receptors. Neuron 21 (2), 393–400. http://dx.doi.org/10.1016/S0896-6273(00)80548-6 (August). Song, Insuk, Kamboj, Sunjeev, Xia, Jun, Dong, Hualing, Liao, Dezhi, Huganir, Richard L., 1998b. Interaction of the N-ethylmaleimide–sensitive factor with AMPA receptors. Neuron 21 (2), 393–400. http://dx.doi.org/10.1016/S0896-6273(00) 80548-6 (August 1). Stegmüller, Judith, Werner, Hauke, Nave, Klaus-Armin, Trotter, Jacqueline, 2003. The proteoglycan NG2 is complexed with alpha-amino-3-hydroxy-5-methyl-4isoxazolepropionic acid (AMPA) receptors by the PDZ glutamate receptor interaction protein (GRIP) in glial progenitor cells. Implications for glial-neuronal signaling. J. Biol. Chem. 278 (6), 3590–3598. http://dx.doi.org/10.1074/jbc.M210010200 (February 7). Stephan, Klaas E., Baldeweg, Torsten, Friston, Karl J., 2006. Synaptic plasticity and dysconnection in schizophrenia. Biol. Psychiatry 59 (10), 929–939 (May 15). Su, Gang, Kuchinsky, Allan, Morris, John H., States, David J., Meng, Fan, 2010. GLay: community structure analysis of biological networks. Bioinforma. (Oxford, England) 26 (24), 3135–3137 (December 15). Subramanian, Aravind, Tamayo, Pablo, Mootha, Vamsi K., Mukherjee, Sayan, Ebert, Benjamin L., Gillette, Michael A., Paulovich, Amanda, et al., 2005. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A. 102 (43), 15545–15550. http://dx.doi.org/ 10.1073/pnas.0506580102 (October 25). Südhof, Thomas C., 2008. Neuroligins and neurexins link synaptic function to cognitive disease. Nature 455 (7215), 903–911 (October 16). Suzuki, H., Fukunishi, Y., Kagawa, I., Saito, R., Oda, H., Endo, T., Kondo, S., Bono, H., Okazaki, Y., Hayashizaki, Y., 2001. Protein–protein interaction panel using mouse full-length cDNAs. Genome Res. 11 (10), 1758–1765. http://dx.doi.org/10.1101/ gr.180101 (October). Takeuchi, Masakazu, Hata, Yutaka, Hirao, Kazuyo, Toyoda, Atsushi, Irie, Mina, Takai, Yoshimi, 11943–11951, 1997. SAPAPs. A family of PSD-95/SAP90-associated proteins localized at postsynaptic density. J. Biol. Chem. 272 (18). http://dx.doi.org/ 10.1074/jbc.272.18.11943 (May).

O

925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981

O. Bar-shira, G. Chechik / Molecular and Cellular Neuroscience xxx (2013) xxx–xxx

E

12

U

N

C

1040

Please cite this article as: Bar-shira, O., Chechik, G., Predicting protein–protein interactions in the post synaptic density, Mol. Cell. Neurosci. (2013), http://dx.doi.org/10.1016/j.mcn.2013.04.004

982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039